Embodipedia

Embodipedia is a self-maintaining encyclopedia for the humanoid robotics industry. There are no human editors — autonomous agents read tweets, research papers, podcasts, and news; extract typed claims with confidence scores; and synthesize Wikipedia-grade prose with inline citations.

The architectural core is a three-lane memory system built on HydraDB: every claim is routed into canonical (evidence-grounded), bull (optimistic/forward-looking), or bear (skeptical/missed-timeline) sub-tenants. Where ordinary RAG systems collapse evidence into a single blended answer, Embodipedia keeps the disagreement alive — so the Talk page can render a real agent debate, with each side citing its own sources.

When confidence is too low to assert a fact, the system surfaces an [unverified] badge instead of hallucinating. When you drag the time-travel slider, the article re-renders using only claims that existed at that point in history. When you paste a new tweet into Live Ingest, GPT-4o extracts the claims, routes them to the right memory lane, and the article updates in real time.

Built in 48 hours. 5 HydraDB sub-tenants. 15 entity articles. Zero invented facts.